# Stochastic Depth

> Source: https://sukruyusufkaya.com/en/glossary/stochastic-depth
> Updated: 2026-05-13T20:02:22.263Z
> Type: glossary
> Category: derin-ogrenme
**TLDR:** A method that provides stronger regularization in very deep networks by randomly skipping some layers during training.

<p>Stochastic depth is used especially in very deep residual architectures to reduce overfitting and optimization difficulty. During training, some blocks are temporarily skipped, while the full network is used at test time. This creates an effect similar to training over multiple depth variants of the model. It can help deep networks become more robust and more generalizable.</p>